MimicGen: A Data Generation System for Scalable Robot Learning using Human Demonstrations

TL;DR

MimicGen is a system that automatically synthesizes large, diverse robot demonstration datasets from few human examples, enabling effective imitation learning for complex tasks without extensive data collection.

Contribution

MimicGen introduces a novel data synthesis approach that scales robot learning datasets from limited human demonstrations, reducing costs and expanding task diversity.

Findings

Generated over 50K demonstrations from ~200 human demos.

Robots trained on MimicGen data perform well in complex tasks.

MimicGen data outperforms additional human demonstrations.

Abstract

Imitation learning from a large set of human demonstrations has proved to be an effective paradigm for building capable robot agents. However, the demonstrations can be extremely costly and time-consuming to collect. We introduce MimicGen, a system for automatically synthesizing large-scale, rich datasets from only a small number of human demonstrations by adapting them to new contexts. We use MimicGen to generate over 50K demonstrations across 18 tasks with diverse scene configurations, object instances, and robot arms from just ~200 human demonstrations. We show that robot agents can be effectively trained on this generated dataset by imitation learning to achieve strong performance in long-horizon and high-precision tasks, such as multi-part assembly and coffee preparation, across broad initial state distributions. We further demonstrate that the effectiveness and utility of MimicGen data compare favorably to collecting additional human demonstrations, making it a powerful and economical approach towards scaling up robot learning. Datasets, simulation environments, videos, and more at https://mimicgen.github.io .